llama.cpp on the M2 Ultra 192GB
On the M2 Ultra 192GB, llama.cpp runs MiniMax M2.7 230B · 10B active at 4-bit at a median 32.7 tokens per second for generation, from 1 plain run (context not stated).
Speed by model on the M2 Ultra 192GB
Plain runs only: one device, one request, no speculative decoding, the whole model in memory. Each row is one model size at one bit class, with the other engines' plain runs of the same model on this card beside it. Context length and build still differ between runs, and each side shows its own.
| Model | llama.cpp | Other engines |
|---|---|---|
| MiniMax M2.7 230B · 10B active 4-bit | 32.7 t/sone report, context not stated | no other engine reported |
| Qwen3.8 27B 6-bit | 22.4 t/sone report, 128K context | no other engine reported |
4 llama.cpp reports on the M2 Ultra 192GB
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- MiniMax M2.7 230B · 10B active · UD-Q4_K_M32.7 t/s
- Mimo 2.5 · UD-Q3_K_M37.9 t/s
- Qwen3.8 27B · Q6_K_XL22.4 t/s
- DeepSeek V4 Flash 284B · 13B active25.8 t/s
Frequently asked
How fast is llama.cpp on the M2 Ultra 192GB?
On the M2 Ultra 192GB, llama.cpp runs MiniMax M2.7 230B · 10B active at 4-bit at a median 32.7 tokens per second for generation, from 1 plain run (context not stated).